SENTIMENT ANALYSIS WITH VALENCY AND TENDENCY FUNCTIONS USING BERT SEQUENCE MODEL

This study explores the potential of analyzing public opinion in Indonesia using advanced deep learning techniques to enhance sentiment analysis. Leveraging the BERT (Bidirectional Encoder Representations from Transformers) sequence model, specifically the BertForSequenceClassification model, we capture nuanced sentiment information through BERT’s deep contextualized word representations. A dual-stage sentiment analysis framework is introduced, integrating the Tendency and Valency models to improve accuracy. The Tendency Model classifies texts into Low Tendency or High Tendency categories, while the Valency Model further refines sentiment analysis within the High Tendency data by evaluating sentiment intensity and distinguishing between positive and negative sentiments. This dual- stage approach significantly outperforms traditional single-stage methods, which achieved a lower accuracy of 35% due to their limited ability to capture nuanced sentiment variations. The dual-stage model demonstrates superior performance, achieving an accuracy of 82% and an F1 score of 78% on test data, indicating high precision in sentiment evaluation. The study highlights the effectiveness of combining deep learning techniques with a dual-stage framework to provide more accurate and contextually aware sentiment classification, advancing the analysis of public opinion with greater precision.

Keywords: Sentiment Analysis, Valency, Tendency, BERT, Deep Learning, Text Classification